Most help center managers would describe their content as "a bit messy" or "in need of a cleanup." What they usually mean is: they haven't looked closely in a while, they suspect there are problems, and addressing those problems sounds like a lot of work.

They're right on all three counts.

The typical Intercom help center accumulates problems the same way a codebase does — incrementally, invisibly, and mostly as a side effect of things going right. You launch a new feature, you write an article. You change a plan, you update the billing page. You get a support ticket three times in a row, you write a FAQ. Over two or three years of this, you end up with 200-plus articles across 40-odd collections, naming conventions that reflect six different eras of how your team thought about documentation, and a handful of articles that were accurate once.

The content isn't wrong, exactly. It just hasn't been looked at as a whole — and looking at it as a whole is precisely the kind of work that never quite makes it onto anyone's priority list.

Until now it was genuinely difficult to do. A thorough audit of a large help center — reading every article, cross-referencing duplicates, checking naming consistency, identifying gaps — is weeks of manual work. Most teams either do it halfway or not at all.

That's changed.

What connecting Claude to Intercom actually makes possible

Claude has a native Intercom integration. When you connect the two, Claude can read your entire published help center programmatically — every article, every collection — and reason about the full body of content at once.

This is meaningfully different from asking an AI to "review my docs" by pasting things in. When Claude has direct API access to your help center, it can:

  • Pull every article in bulk, not just the ones you thought to share
  • Compare article content across your entire library to find overlaps you wouldn't have noticed
  • Trace structural patterns — naming conventions, collection organization, cross-linking — across hundreds of articles simultaneously
  • Flag content that references specific dates, deprecated features, or old plan names without you having to manually identify them first

The result is an audit that's comprehensive in a way that human audits rarely are, produced in a fraction of the time.

What a full audit actually covers

A properly scoped help center audit has six distinct areas — and most teams, when they do audits at all, tend to focus on only one or two.

Knowledge gaps are the hardest to see from inside the content. You're looking for features mentioned in passing that never got their own article, integrations with a setup guide but no troubleshooting companion, and the implicit "next step" questions that articles leave open. These gaps are often the ones generating your most repetitive support tickets.

Duplicate and overlapping content is more common than most teams expect. FAQ articles that substantially restate a how-to guide. Two articles covering the same feature from slightly different angles. A "how it works" piece that covers 80% of the same ground as the setup guide next to it. Duplication isn't just a maintenance burden — it creates real confusion for users who aren't sure which article to trust.

Naming convention inconsistencies are a proxy for organizational debt. If your help center has grown across a few years and a few different team members, you'll almost certainly have setup guides titled "Setup Guide: [Tool]" alongside others titled "How to connect [Tool]" alongside others with no type prefix at all. Users can't tell what kind of article they're clicking into, and your team can't efficiently maintain content when there's no consistent pattern.

Structural issues — thin collections, misplaced content, collections whose scope is unclear from their titles — affect discoverability more than most people realize. A collection with one article isn't a collection; it's a navigation dead end.

Stale content is the most urgent category when it surfaces. Beta labels that were never removed. Articles referencing pricing plans that were updated eighteen months ago. Date-specific copy that silently became inaccurate. This is the content that actively misleads users.

Missing cross-links are the quietest problem. Two articles that clearly belong together — a setup guide and its troubleshooting companion, a billing article and the plan comparison page — but don't reference each other. Every missing cross-link is a navigation path that forces users into your search bar or your support queue.

What you get out the other end

The output is a structured report, not a vague set of observations. Each issue includes the specific article IDs involved, a description of the problem, and an actionable recommendation. The closing priority matrix tells you what to fix first if you're resource-constrained — and most teams are.

For a help center of 150-300 articles, a full audit typically surfaces somewhere between 30 and 60 discrete issues. Not all of them matter equally. But the ten that sit at the intersection of high value and low effort — the ones that show up in the priority list — are usually fixable in an afternoon.

That's the point. Not to generate a document that sits in a folder. To produce a clear, prioritized list of changes that a real person can actually work through.

Getting started

You'll need Claude.ai with the Intercom integration connected. Once that's in place, copy the prompt below, swap in your own help center details — company name, article count, topic areas, known naming patterns — and send it. Claude will handle the rest.

The whole process — from running the audit to having a structured report — takes less time than a single editorial planning meeting.

Your help center has probably been "a bit messy" for long enough.

Prompt:

You are a knowledge base analyst auditing the [Your company] Intercom help center. There are approximately [N] published articles across [N] collections. Your job is to produce a structured audit report.

You have access to the Intercom MCP connector. Use list_articles (per_page: 150, page: 1 then page: 2 if needed) to fetch all articles, then use get_article to retrieve full content for articles that need closer review. Work through the complete set before drawing conclusions — do not summarize based on titles alone.

---

CONTEXT

[Your company] is a [your product description]. The help center covers: [list your help center topic areas].

---

AUDIT SCOPE

1. KNOWLEDGE GAPS
Identify topics or workflows that are missing or inadequately covered:
- Are there product features mentioned in passing within articles but never fully documented?
- Are there logical "next step" questions left unanswered at the end of articles?
- Are there integrations or features with a setup guide but no FAQ, troubleshooting article, or companion "how it works" piece?
- Are there any program-specific areas (partner programs, enterprise features, API docs) that are underdocumented relative to their likely usage?

2. DUPLICATE OR OVERLAPPING CONTENT
Flag articles whose scope significantly overlaps with another:
- Identify any FAQ article that substantially duplicates a dedicated how-to article on the same topic.
- Flag any article pairs where the content could be merged or where one makes the other redundant.
- Note article IDs and titles for each flagged pair.

3. NAMING CONVENTION INCONSISTENCIES
Identify inconsistencies within each article type pattern you observe:
- Setup guides, FAQs, troubleshooting articles, playbooks, best practices guides — note what the dominant pattern is and which articles deviate.
- Propose a single canonical naming convention for each article type found.

4. STRUCTURAL ISSUES
- List any collections containing only 1 article. Recommend whether each should be merged into a broader collection or kept separate.
- Note any collections whose name or scope is unclear from the article titles alone.
- Recommend whether integration-specific content should be consolidated per integration, or organized by content type across all integrations.

5. STALE OR TIME-SENSITIVE CONTENT
Flag articles that may need a freshness review:
- Any article with "(Beta)" in the title or that describes a feature as in-progress
- Any article referencing specific dates, promotions, or limited-time features
- Any article describing pricing, plans, or billing terms that may have changed

6. MISSING CROSS-LINKS
For article pairs that are clearly related but likely don't reference each other, flag them and suggest where a "See also" or inline link would improve navigation. Focus on:
- Setup guides that don't link to their companion FAQ or troubleshooting article
- Billing/plan articles that don't cross-reference each other
- Any feature area where a user reading one article would naturally need to find another

---

OUTPUT FORMAT

Return a structured report with one section per audit area. For each issue include:
- Article ID(s) and title(s)
- One-sentence description of the problem
- A concrete recommendation

Close with a prioritized action list: the 10 highest-impact changes, ranked by estimated effort (low / medium / high) and value (low / medium / high).